Types of studies

We will include randomized controlled trials (RCTs), clinical controlled trials (CCTs), cluster‐randomized trials, and non‐randomized studies with a control or comparison group.

We will also include specific non‐randomized studies (e.g. case‐control, longitudinal cohorts) to complement data that we gather from RCTs, such as baseline risk estimates and overall effectiveness of food‐based calcium and vitamin D intake. Non‐randomized studies can also provide sequential evidence about the longer‐term effects that these food‐based nutrients have on bone health, which may not be observed in RCTs with shorter time frames. The included study designs are based on the Effective Practice and Organization of Care (EPOC) Guidelines.

1. Controlled before‐and‐after studies (CBAs), provided that:

   1. the data in the experimental and control sites were collected in the same time frame, and/or

   2. in the case where studies use a second site as a comparison, the study and control sites were comparable with respect to setting and population.

2. Interrupted time series (ITS) studies (with at least three measurement points before and after intervention), provided the study reported that the intervention occurred at a clearly defined point in time, the study recorded three or more data points at the beginning and at the end of the intervention, and if a repeated measure analysis was performed by investigators. We will exclude ITS studies that ignored secular (trend) changes and that performed a simple t‐test of the pre‐ versus post‐intervention periods without providing justification.

Types of participants

We will include trials that enrol postmenopausal women of any age, where menopause is defined by trial authors (typically defined as 12 months since the cessation of menstruation) (Harlow 2012). Studies including postmenopausal women with or without history of any fracture will be included. Faced with instances where study data on postmenopausal women are combined with premenopausal women, the authors will be contacted for disaggregated data. The study will be excluded from data analysis if no response is received after three consecutive attempts at contact. Within both vertebral and non‐vertebral fractures, women who have had traumatic fractures will be combined with women without a history of fractures. Clinical conditions outside of osteoporosis (e.g. chronic kidney disease) as reported by study authors, or known medications (e.g. aromatase inhibitors) that impact bone metabolism will be considered as confounders.

Types of interventions

We will include RCTs and CCTs that examine calcium or vitamin D foods, or both, versus comparison diets (e.g. usual diet, food counselling only) in this review. Participants in the intervention groups will receive high calcium or vitamin D foods, or both, as defined by the authors. This may include high‐content calcium and/or vitamin D foods determined using food lists developed by the United States Department of Agriculture Dietary Guidelines for Americans 2020 as a reference (USDA 2020). Food counseling in the context of this review is defined as counseling for intake of foods high in calcium or in vitamin D, or both. We will also consider studies in which food‐based calcium or vitamin D or both were compared to non‐food forms of calcium or vitamin D or both (i.e. supplements) groups.

In addition to diet‐only interventions, we will include trials that provide food‐based calcium or vitamin D, or both, as part of a multi‐component intervention (e.g. provision of high calcium and vitamin D foods with counseling and physical activity). The effect of physical activity alone in osteoporosis has been explored by a previous Cochrane Review (Howe 2011) and therefore will not be considered in this review. We will not include studies that examine the effects of calcium and/or vitamin D alone since previous Cochrane Reviews have examined this research question.

The comparisons of interest are.

• Provision of high‐content calcium or vitamin D foods, or both, versus supplement, usual diet, or food counseling only.

• Provision of high‐content calcium or vitamin D foods, or both, and food counseling versus supplement, usual diet, or food counseling only.

• Provision of high calcium and vitamin D foods with counseling and physical activity versus physical activity only.

Types of outcome measures

Electronic searches

We will perform searches in the following electronic databases:

• Cochrane Central Register of Controlled Trials (CENTRAL) in the Cochrane Library (current issue);

• MEDLINE via OVID (1948 to present);

• Embase via Ovid (1974 to current);

• CINAHL (Cumulative Index to Nursing and Allied Health Literature) via EBSCOhost (1982 to present);

• AGRICOLA via OVID (1970 to present);

• Food Science Technology Abstracts via OVID (1969 to present);

• OLDMEDLINE;

• PsycLIT.

The search strategy described in Appendix 1 for MEDLINE will be applied and adopted for similar electronic databases, if appropriate. Databases will be searched from their inception to the present day and will not be restricted by the language of publication. Additionally, we will screen upcoming trials registered at the US National Institutes of Health Ongoing Trials Register ClinicalTrials.gov (www.ClinicalTrials.gov) and the World Health Organization trials portal (www.who.int/ictrp/en/). Studies in the published and unpublished or gray literature will also be examined for this review.

Searching other resources

We will review the reference lists of all primary studies, as well as identified trials for additional references. Relevant manufacturers' websites will be searched for information on trials with a supplement arm that may also include a food‐based intervention arm. Additionally, we will search for errata or retractions from included studies published in full text on PubMed (pubmed.ncbi.nlm.nih.gov) and report the dates that these searches were completed in the review.

Selection of studies

Four review authors, in groups of two (Damian K Francis (DKF) and Annie W Lin (AWL); Beatrice J Leyaro (BJL) and Maduka de Lanerolle Dias (MLD)), will independently screen titles and abstracts that are identified as part of the search. Studies will be coded according to the review inclusion criteria as 'retrieve' (eligible or potentially eligible/unclear) or 'do not retrieve'. Full‐text reports/publications will be retrieved for studies coded as 'retrieve.' In the same groups of two (DKF and AWL; BJL and MLD), the review authors will independently screen the full‐text reports/publications, identify any studies for inclusion, and record reasons for exclusion of any ineligible studies. The screening process will be managed using Covidence (Covidence). At each step, the four review authors (DKF, AWL, BJL, MLD) will resolve any disagreements through discussion or, if required, through consultation with a third author (DKF or Patricia A Cassano (PAC)). We will identify and exclude duplicates and collate multiple reports of the same study so that each study, rather than each report, is the unit of interest in the review. We will record the selection process in sufficient detail to complete a PRISMA flow diagram (PRISMA Group 2009) and 'Characteristics of excluded studies' table.

Data extraction and management

We will use a data collection form for study characteristics and outcome data, which has been pilot tested on at least one study to be included in the review. Four review authors, in groups of two (AWL and Brittany Y Jarrett (BYJ); BJL and MLD) will conduct duplicate data extraction for relevant study characteristics (i.e. methods, participants, interventions, outcome data). As outlined in the data extraction form (Appendix 2), we will also extract equity data using the PROGRESS guidelines (Mbuagbaw 2017).

We will extract the number of events and sample size per treatment group for dichotomous outcomes, and means, standard deviations, and sample size per treatment group for continuous outcomes. Baseline characteristics data such as calcium and vitamin D intake and blood levels as well as adherence to intervention will also be extracted. We will note in the 'Characteristics of included studies' table when outcome data were not reported in a usable way or when outcome data were transformed or estimated from a study graph, or both. We will resolve any disagreements (i.e. within and between each group of two) by consensus or by involving a third author from the data extraction team (DKF, AWL, BJL, MLD). We will contact the corresponding author of an original report when it is not possible to obtain data from the study graphs, or when there are data extraction questions that cannot be reliably answered using the published data. Medications known to impact bone metabolism, use of anti‐resorptive therapy, body mass index, obesity, diabetes, tobacco smoking, and age over 65 years (Looker 2015) will be considered as confounders.

Two review authors (DKF and AWL) will transfer the outcome data into the Review Manager 5.4 (Review Manager 2020) for analysis. We will double‐check that data have been entered correctly by comparing the data presented in the systematic review with the original study reports.

For RCTs that present pre‐ (i.e. baseline) and post‐intervention values, as well as change data, we will extract the change from the study report. If change has not been reported, then we will calculate change and an appropriate measure of dispersion based on guidance from the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2021). For other study designs, we will extract the post‐intervention values, as appropriate, under the assumption that randomization leads to groups that are similar on pre‐intervention (i.e. baseline) values. We will extract unadjusted values for RCTs and adjusted values for non‐randomized studies of interventions (NRSIs) as per the Cochrane Handbook for Systematic Reviews of Interventions (Reeves 2021). If the studies adjust for different factors, then we will only include unadjusted values in the subsequent meta‐analysis, or we will perform sensitivity analyses to address whether intervention effects vary by covariates included in models. We will extract outcome data that were analyzed in the intention‐to‐treat (ITT) analysis. If these types of outcome data are not available, then we will extract values per protocol or as treated, depending on availability.

Assessment of risk of bias in included studies

Four review authors (DKF, AWL, BJL, MLD) will independently assess the risk of bias for each study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021a) and as outlined in Appendix 3. We will assess the risk of bias for included studies using Cochrane risk of bias 2 (RoB 2) tool and ROBINS‐I (Risk Of Bias In Non‐randomized Studies ‐ of Interventions) tool. Disagreements in assessing risk of bias will be resolved by involving a fifth author (PAC). 

We will assess the effect of intention‐treat analysis as well as adhering to intervention as specified in the trial protocol. For 'per‐protocol effect' the following deviations from the intended food‐based calcium intervention will be examined:

• occurrence of non‐protocol interventions (e.g. use of calcium supplement);

• non‐adherence by trial participants to their assigned intervention (concomitant use of calcium supplements);

• failures in implementing the intervention that could affect the outcome (e.g. inadequate vitamin D intake).

Risk of bias assessment will be done for major outcomes; that is, vertebral and non‐vertebral fractures, quality of life, adverse effects, and withdrawals due to adverse effects. We will consider similarity of outcome measure across outcomes such as MRI to detect fractures, and at similar time points of 3 months or more post‐intervention. 

The domains included in RoB 2 are in Additional Table 1. Using signaling questions, judgment in each domain will be expressed as 'low risk of bias', 'some concerns', or 'high risk of bias'. Decisions to arrive at these judgments will be guided by algorithms that map responses to signaling questions to a proposed risk of bias judgment at www.riskofbias.info. The process of managing assessment of risk of bias will be done using the RoB 2 Excel tool available on www.riskofbias.info. 

A key difference between cluster‐randomized trials and individually‐randomized trials is that participants may be recruited to the trial or otherwise identified for inclusion in the analysis after the clusters have been randomized. This can lead to bias if knowledge of the intervention assigned to a cluster affects recruitment or identification of participants (Higgins 2021b). As such, we will follow guidance for assessing risk of bias for cluster‐randomized trials as outlined by Higgins 2021b.  

For NRSIs with a control or comparator group, including controlled before‐and‐after and follow‐up studies, we will use the validated ROBINS‐I tool to assess risk of bias (Sterne 2016). In each domain, we will assign a judgment of 'low risk of bias', 'moderate risk of bias', 'serious risk of bias', 'critical risk of bias', and 'no information' (Sterne 2016).

We will implement ROBINS‐I (Additional Table 2) in a six‐step process which includes:

1. specify the research question through consideration of a target trial;

2. specify the outcome and result being assessed;

3. for the specified result, examine how the confounders (e.g. conditions outside osteoporosis and medications affecting bone metabolism) and co‐interventions were addressed;

4. answer signaling questions for the seven bias domains;

5. formulate risk of bias judgments for each of the seven bias domains, informed by answers to the signaling questions;

6. formulate an overall judgment on risk of bias for the outcome and result being assessed.

The ROBINS‐I assesses risk of bias at three levels (using seven domains with appropriate signaling questions), namely:

1. pre‐intervention;

2. at intervention, and

3. post‐intervention.

The review authors will not be blinded to the names of study authors, institutions, journals, and results of included studies. If there are concerns or issues, they will be resolved by discussion or with the opinion of a fifth review author (PAC). Results will be recorded in the relevant 'Characteristics of included studies' tables in Review Manager (Review Manager 2020) and summarized in a risk of bias table or graph.

Measures of treatment effect

Unit of analysis issues

Dealing with missing data

We will contact investigators or study sponsors to obtain missing numerical outcome data where needed (e.g. when a study is identified as abstract only or when data are not available for all participants). In situations where this is not possible, and the missing data are thought to introduce serious bias, we will investigate the impact of including these studies in the overall assessment of results by performing a sensitivity analysis. Any assumptions and imputations to handle missing data will be clearly described and the effect of imputation will be explored (Deeks 2021).

For dichotomous outcomes (e.g. diagnosis of osteoporosis), the diagnosis rate (incidence) will be calculated using the number of patients randomized in the group as the denominator. For continuous outcomes (e.g. mean change in BMD), we will calculate the MD or SMD based on the number of patients analyzed at a particular time point. If the number of patients analyzed is not presented for each time point, then the number of randomized patients in each group at baseline will be used. Where possible, missing standard deviations will be computed from other statistics such as standard errors, CIs or P values, according to the methods recommended in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021c). The potential effect of missing data on conclusions drawn from this review will be described or considered, or both, in the sensitivity analyses.

Assessment of heterogeneity

Clinical and methodological diversity will be assessed in terms of participants, interventions, outcomes, and study characteristics for the included studies to determine whether a meta‐analysis is appropriate. Clinical homogeneity will be examined with respect to participants, interventions, and outcome measures from the data extraction tables. Statistical heterogeneity will be assessed by visual inspection of the forest plot to assess obvious differences in results between the studies and by using the I² and Chi² statistics. We will use the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2021) recommendations to interpret the I² value:

• 0% to 40% might 'not be important';

• 30% to 60% may represent 'moderate' heterogeneity

• 50% to 90% may represent 'substantial' heterogeneity;

• 75% to 100% represents 'considerable' heterogeneity.

As noted in the Cochrane Handbook for Systematic Reviews of Interventions, we will keep in mind that the importance of I² depends on the magnitude and direction of effects and the strength of evidence for heterogeneity. The Chi² test will be interpreted where a P value ≤ 0.10 will indicate evidence of statistical heterogeneity. If we identify substantial heterogeneity, then we will report it and investigate its possible causes though a priori and post hoc subgroup analyses and meta‐regression.

Assessment of reporting biases

We will perform a comparison‐adjusted funnel plot for each outcome (when ≥ 10 studies are available) to explore possible study biases. We will undertake formal statistical tests to investigate funnel plot asymmetry by conducting a linear regression of the effect estimate of the calcium or vitamin D, or both, against the standard error, of the estimate weighted by the inverse of the variance of the intervention (Egger 1997). To assess outcome reporting bias, we will check trial protocols against published reports. For studies published after July 2005, we will screen the clinical trial register at the International Clinical Trials Registry Platform of the World Health Organization (trialsearch.who.int) for the a priori trial protocol. We will evaluate whether selective reporting of outcomes is present.

Data synthesis

Subgroup analysis and investigation of heterogeneity

We plan to carry out the following subgroup analyses for vertebral and non‐vertebral fractures.

• Duration of the intervention: 3 to less than 6 months; 6 months to less than 1 year; 1 year or longer.

• Age: 64 years and younger; 65 years and older.

• Proportion of dietary reference intake of calcium and vitamin D provided or recommended by the Institute of Medicine: less than estimated average requirements (EAR); greater or equal to EAR and less than recommended daily allowance (RDA); greater or equal to RDA.

• Baseline calcium and/or vitamin D status following values established by the Institute of Medicine (IOM 2011).

• History of fragility fractures among postmenopausal women.

• Vitamin D insufficiency (< 30 ng/dL) or inadequate intake of calcium using recommended RDA by the Institute of Medicine (IOM 2011).

We will use the formal test for subgroup interactions in Review Manager (Review Manager 2020) and will use caution in the interpretation of subgroup analyses as advised in Section 10 of the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2021). The magnitude of the effects will be compared between the subgroups by means of assessing the overlap of the CIs of the summary estimate.

Sensitivity analysis

To investigate the robustness of the treatment effect of calcium or vitamin D foods, or both, on vertebral and non‐vertebral fractures, the following sensitivity analyses will be conducted.

1. Including all eligible studies, including those with high risk of overall bias.

2. Omitting studies with missing or imputed data.

3. The impact of different ICCs for cluster‐randomized trials on the results (if these are included for RCTs).

Summary of findings and assessment of the certainty of the evidence

We will create a summary of findings (SoF) using GRADEpro GDT software (GRADEpro GDT 2021). The template summary tables will include information on the following (Additional Table 3):

• comparisons of interest;

• study design (population, intervention, comparison, outcome (PICO));

• location;

• time point;

• outcome measurements.

In situations where there is RCT and non‐RCT evidence for the outcome, we will report findings from the RCT. Additionally, for each major outcome, we will include the evidence certainty rating, as assessed through the GRADE approach (GRADEpro GDT 2021). We will include the following outcomes in the SoFs table.

1. Vertebral fractures.

2. Non‐vertebral fractures (i.e. hip, wrist, forearms).

3. Quality of life.

4. Withdrawals due to adverse events.

5. Serious adverse events (i.e. hospitalizations, events resulting in disability or death).

6. Number of patients experiencing any adverse events.

7. Bone mineral density.

When assessing the certainty of the evidence, the overall RoB 2 judgment will be used to feed into the GRADE assessment. We will refer to Chapters 8, 14, and 15 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2021a; Schünemann 2021a; Schünemann 2021b). The five GRADE considerations (i.e. study limitations, consistency of effect, imprecision, indirectness, and publication bias) will be used to evaluate the certainty of a body of evidence as high, moderate, low, or very low. We will also consider the following criteria in upgrading the certainty of the evidence when appropriate: large effect size, dose‐response gradient, and plausible confounding effect. We will use GRADEpro software to prepare the SoF tables using the version 3 SoF for display (GRADEpro GDT 2021). Through the use of footnotes, we will provide justification for all decisions to down‐ or upgrade the quality of the studies.